I have a complex question.

Is there any other way to model the preference misalignments of the sendres (in senders-receivers games) towards the desire of some specific state apart from the quadratic loss functions? Why not quadratic gains functions? I understand that it is a bit intuitive to use this model, but is there any other model that is also used for such situations?


1 Answer 1


Quadratic loss is convenient because it provides a close form solution after investigating the first order conditions.

That said, if you're interested in belief updating, an absolute loss function can modify (slow down) the update when a signal happens to be a the tail of the distribution and look like an outlier.

You might also consider a composite function $L$ (for example a weighted average of a quadratic loss and absolute loss) to control how fast those updates are made. Let $x^*$ be the true value, and $x$ an estimate $$L_{x^*}(x)=\alpha \left( x^* - x \right)^2 + \beta \left| x^* - x \right| $$ where $\alpha$ and $\beta$ could be any non-negative real number (e.g. $\alpha=\beta=0.5$).

  • $\begingroup$ A weighted average of a quadratic loss and absolute loss? How would it look like? Could you write down the formula that you have in mind? $\endgroup$ Commented Dec 31, 2023 at 10:37
  • 1
    $\begingroup$ It's done, please let me know if you have further questions. $\endgroup$ Commented Dec 31, 2023 at 22:56
  • $\begingroup$ Thank you! I will check this out! $\endgroup$ Commented Dec 31, 2023 at 22:57

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service and acknowledge you have read our privacy policy.

Not the answer you're looking for? Browse other questions tagged or ask your own question.